Congressional District, Political Leaning, and School Funding

Setting the Stage

Over this past year, states and local governments have begun a one-a-decade process of drawing new voting district boundaries. This redistricting uses population and demographic data from the census conducted by the Census Bureau, with the most recent occurring in 2020.

The redistribution process was created to accurately reflect how communities change over time in the US legislative system. Maps are redrawn to ensure that districts are equally populated and representative of the district’s population. When done correctly and equitably, redistricting is a window for the political electoral processes to adapt to constituent demographics accurately, ensuring that constituents are accurately represented in local, state, and national legislative. In the words of John Adams, “an exact portrait, a miniature” of the people and communities grouped geographically,

Yet, redistricting has become the most pressing threat to democracy. As described by the Brennan Center:

But sometimes the process is used to draw maps that put a thumb on the scale to manufacture election outcomes that are detached from the preferences of voters. Rather than voters choosing their representatives, gerrymandering empowers politicians to choose their voters. This tends to occur especially when linedrawing is left to legislatures and one political party controls the process, as has become increasingly common. When that happens, partisan concerns almost invariably take precedence over all else. That produces maps where electoral results are virtually guaranteed even in years where the party drawing maps has a bad year.

Alongside a 2019 Supreme Court ruling that prohibits redistricting to be challenged in federal court, this reaffirms major threats to health, education, and social safety net policies at state and national levels, even moreso heightened midst a time of deep divisive partisanship in the United States.

How Gerrymadering Manipulates District Boundaries Image source

In summary,

  • Rigged maps make elections less competitive, severly dampening the voices and political voices of all Americans

  • Both historical and current residential segregation means that marginalized communities are the targets of pointed redistricting; effectively creating a tool for parties to slice and dice districts in ways that discriminate against communities of color

  • Gerrymandering is getting worse. Applying data science to work, intricate computer algorithms and sophisticated date on voter patterns have allowed political gamesmanship to “maximize” partisan benefit

Intial Questions

As a part of a larger exploration of the impact of educational and social safety net policies and sociodemographic characteristics at the state and school district levels on predictors of student success. We apply the SEDA data set to provide a comprehensive picture of factors at the family, district, and state-level factors that we hypothesize influence student educational outcomes.

To prime further data analyses, the objective of this first portion is to investigate the differences, if any, in school district expenditure between political preferences at the Congressional District level.

Specifically,

Does spending on school districts differ by political ideology?

The analyses below highlight differences by drawing from data on per student expenditure and per teacher expenditures. This was facilitated at the Congressional District level to capture the granularities unafforded when looking at state-only levels.

An interactive map facilitates this comparison by grouping dominant political preferences and expenditure per student and teacher averaged over school districts in a Congressional District.

As mentioned in class,

Plots of maps can be very powerful, very informative and very aesthetically pleasing visualizations.

The goal was to be able to provide a compelling story using nationally representative data.

All in all, we hope that this work underscores the potential impact of gerrymandering on students and teachers by visualizing differences in spending in education for grades 3-8 based on the dominant political ideology in Congressional districts.

Data Sources

School Expenditure data are available from: https://exhibits.stanford.edu/data/catalog/db586ns4974

Political Ideology data are available from: https://americanideologyproject.com/

Variables of interest include:

Background: Party Platform Positions on Education

This question arose as way to compare the two major parties – Democrats and Republicans – to illuminate how political platforms and values have direct impacts at the level of school districts.

An overview of the party education platforms is included below:

Comparison between Dem and GOP platforms for K-12 education
Topic Democrat Republican
Educators
  • Public school educators should have the same rights to organize, join a union, and collectively bargain as private sector workers
  • Teachers should be protected against “frivolous” lawsuits and should be able to take reasonable actions to maintain discipline and order in the classrooms
  • Fight to increase pay and benefits for all educators, in order to help recruit, retain, and reward high-quality teachers and support professionals, and will continue to support measures to help teachers pay for out-of-pocket classroom expenses
  • Urge school districts to make use of teaching talent in the business community, STEM fields, and the military
  • Support programs to enable support staff to climb the professional ladder within schools and recruit a diverse educational workforce
  • Rigid tenure systems should be replaced with merit-based appraoches
Charter Schools
  • Students should have multiple public school pathways to be able to access the opportunities they deserve
  • Education savings accounts, vouchers, and tuition tax credits
  • In support of magnet schools, IB programs, early college high schools
  • Ban for for-proft private charter businesses from receiving federal funding
  • Charter schools subject to the same accountability as public schools (including racial equity, admissions practices, disciplinary procedures, and school finances
  • Oppose private school vouches and other policies that divert taxpayer-funded resources away from public school system
Curriculum and Instruction
  • STEAM education for underrepresented students
  • Opposed to Common Core, and congratulate the states which have successfully repealed it
  • Good understanding of the Bible “being indispensable for the development of an educated citizenry, we encourage state legislatures to offer the Bible in a literature curriculum as an elective in America’s high schools
  • Replace family planning for teens with sexual risk avoidance education that sets abstinence until marriage as the responsible and respected standard of behavior
Funding
  • Closing the school funding gap by tripling Title I funding (benefits low-income students and incentivize states to adopt progressive funding formulas that direct resources to the schools that need it most
  • More money does not necessarily equal better performance (results do not justify spending levels)
  • Funding for rural schools, schools in impoverished urban areas impacted by gentrification, and Bureau of Indian Education schools
  • Increased support for English learners
  • Fully fund the Individuals with Disabilities Education Act for a free appropriate public education and least restrictive environment
Source:
https://www.ascd.org/blogs/what-the-democrats-and-republicans-stand-for-on-education-excerpts-from-party-platforms

Exploratory Analyses: Mapping Constituent Ideology by Congressional District

To begin, we first must accurately delineate between political right and left Congressional districts.

Here, we applied estimates of political preferences from a collection of survey measures to capture the distribution of political preferences based on those living in an geographic area.

This was purposeful. As mentioned above, one concern we had was that, as a result of gerrymandering, relying only on the election results and the parties of Congressional representatives may underweight the true distribution of political preferences among those living in the District. Therefore, we drew from an opensource project, The American Ideology project, (Tausanovitch and Warshaw, 2013)[https://americanideologyproject.com/], to:

  1. explore the distribution of political preferences by geographic regions; and
  2. the magnitude, or “how radical” preferences are.
# Loading USAboundaries from GitHub
devtools::install_github("ropensci/USAboundaries")
devtools::install_github("ropensci/USAboundariesData")

# Using USAboundaries package, load the spatial data 2010-2020 Congressional districts
spatial <- us_congressional()


# Loading congressional ideology data for the 113th Congress from The American Ideology Project, https://americanideologyproject.com/
ideology <- read.csv('cd_113_TW_ideology_estimates.csv')


# Joining political ideology data with the spatial data
## String processing spatial dataset to remove leading zeros in "geoid"
spatial <- spatial %>% # Removing Alaska, Hawaii, and Puerto Rico
  mutate(cd_fips = str_remove(spatial$geoid, "^0+")) %>% 
  filter(state_abbr != 'AK') %>%
  filter(state_abbr != 'HI') %>%
  filter(state_abbr != 'PR')

## Selecting columns of interest from ideology dataset
ideology <- ideology %>%
  select(c('cd_fips', 'mrp_mean', 'pres_2012')) %>%
  mutate(cd_fips = as.character(cd_fips))

map <- left_join(spatial, ideology, by = 'cd_fips')

# Mapping
ideology_map <- ggplot() +
  geom_sf(data = map, 
          aes(geometry=geometry, fill=mrp_mean), alpha=2) +
  scale_fill_distiller(palette='RdBu', 
                       limits = c(-1.2,1.2), 
                       direction = -1, 
                       breaks=c(-1,0,1),
                       labels=c('Left','Center','Right'),
                       name='Political Leaning') +
  theme(panel.grid.major = element_blank(), 
        panel.background = element_blank(),
        axis.title.x=element_blank(),
        axis.text.x=element_blank(),
        axis.ticks.x=element_blank(),
        axis.title.y=element_blank(),
        axis.text.y=element_blank(),
        axis.ticks.y=element_blank(),
        plot.subtitle = element_text(size=9, color='grey45')) +
  labs(title='Contituent Political Ideology by Congressional District', 
       subtitle = 'Source: The American Ideology Project (Tausanovitch & Warshaw, 2015)')

ideology_map

As shown, we do find spatial correlations in political preferences. Politically right Congressional districts are clustered in the Midwest and South, and politically left Congressional districts tend to be located on the coasts.

In addition to, we find that the extent to which these districts are right or left leaning is not balanced. Districts that are more ive (or right-leaning), on average, are much greater than liberal (or left-leaning) Districts. In other words, ive Districts tend to be much more radically so than liberal Districts.

Now, how does this relate to spending on education?

SEDA Dataset Prep

For the following code, please ensure that you have the Stanford Education Data Archive codebook_covariates_v1_1.xlsx installed and available in your working directory

To clean the data, we collapsed school districts into their respective Congressional Districts, averaging the per pupil expenditure and per instructor expenditure for a single measure per Congressional District.

# Loading SEDA dataset
full_data_yr <- read_dta("district by year by grade covariates from acs and ccd master_v1_1.dta")

# New data frame of variables of interest
dat <- full_data_yr %>% # filtering to 2013
  filter(year ==2013) 

dat <- dat[!duplicated(dat$leaid),] %>% # keeping only one observation per school district (removing grade level layer)
  select('leaid','cdcode','ppexp_tot','ppexp_inst')  

dat_na <- sum(is.na(dat$ppexp_tot), is.na(dat$ppexp_inst)) # any NAs?  
x <- is.na(dat$ppexp_inst) 
y <- is.na(dat$ppexp_tot)
nonpaired_na <- sum(any(x != y)) # Confirms that all NA values are paired, so we can remove rows for both ppexp_tot and ppexp_inst combined 
rm(x,y,nonpaired_na, dat_na)

dat <- dat[!(is.na(dat$ppexp_tot)), ] # removing 50 NA values (25 observations)

# Final clean dataset with avg values by Congressional district
dat <- ddply(dat, ~cdcode, transform, ppexp=mean(ppexp_tot),  piexp=mean(ppexp_inst))

seda <- dat[!duplicated(dat$cdcode),] %>% # keeping only one observation per Congressional district
  select(-'leaid',-'ppexp_tot',-'ppexp_inst') # dropping school district measures

Exploratory Analysis: Distribution of Education Funding

Now that we’ve delineated between political right and left Congressional District, we explored how per student and per teacher spending differs between Districts.

A layered histogram will suffice. From this we can explore:

  1. How much is invested in students and teachers, scaled using the underlying dispersion of values (in units of standard deviation)

  2. How clustered is the spending? i.e. plot skew and spread

But first, we want to compare spending based on the national mean and variance. This provides a way to accurately adjust how we can analyze where Congressional districts fall related to spending across all districts.

To do such, a scale (standard deviations) was applied to per student and teacher expenditures in the remaining analyses.

# Joining education dataset with map dataset
seda <- seda %>% 
  mutate(cd_fips=as.character(cdcode)) %>%
  select(-'cdcode')

map <- left_join(map, seda, by = 'cd_fips') 

# Per pupil Congressional district expenditure
map$z <- scale(map$ppexp)

# Per instructor Congressional district Expenditure
map$z_i <- scale(map$piexp)

# Creating indicator variable for histogram layers
hist <- map %>% 
  mutate(left = ifelse(mrp_mean>0, 'Right', 'Left')) # indicator variable for right (1) and left (0)

hist_na <- sum(is.na(hist$left)) # any NAs for unknown political ideology? 
hist <- hist[!(is.na(hist$left)), ] # removing 1 NA value
rm(hist_na)

# Distribution of per Pupil expenditures by Political Ideology
hist_p <- ggplot(hist, aes(x=z,color=left, fill=left)) +
  geom_histogram(position='identity', alpha = 0.3) +
  scale_color_manual(values=c('steelblue','salmon')) +
  scale_fill_manual(values=c('steelblue','salmon')) +
  labs(title='Distribution of Per Pupil Expediture',
       x='Scaled (standard deviation units)',
       y='Count',
       fill="Political Leaning") +
  guides(color=F) +
  theme_light()

# Distribution of per Instructor expenditures by Political Ideology
hist_i <- ggplot(hist, aes(x=z_i,color=left, fill=left)) +
  geom_histogram(position='identity', alpha = 0.3) +
  scale_color_manual(values=c('steelblue','salmon')) +
  scale_fill_manual(values=c('steelblue','salmon')) +
  labs(title='Distribution of Per Instructor Expediture',
       x='Scaled (standard deviation units)',
       y='Count',
       fill="Political Leaning") +
  guides(color=F) +
  theme_light()

## Dropping NAs to calculate group averages
z_avgs <- hist[complete.cases(hist$z),] 
z_i_avgs <- hist[complete.cases(hist$z_i),]
na_z <- sum(is.na(hist$z)) # Ensuring that NAs are properly dropped
na_z_i <- sum(is.na(hist$z_i)) # Ensuring that NAs are properly dropped
rm(na_z, na_z_i)

## Data frame with group averages
mu <- ddply(z_avgs, 'left', summarize, grp.mean=mean(z))
mu_i <- ddply(z_i_avgs, 'left', summarize, grp.mean=mean(z_i))

# Final histogram of per Pupil expenditures by Political Ideology
hist_p <- hist_p + geom_vline(xintercept=mu$grp.mean[1], color='dodgerblue', linetype='dashed') +
  geom_vline(xintercept=mu$grp.mean[2], color='tomato', linetype='longdash'); hist_p

# Final histogram of per Instructor expenditures by Political Ideology
hist_i <- hist_i + geom_vline(xintercept=mu_i$grp.mean[1], color='dodgerblue', linetype='dashed') +
  geom_vline(xintercept=mu_i$grp.mean[2], color='tomato', linetype='longdash'); hist_i

From the histogram, we find that, indeed, confirms our hypothesis: that, as expected from in party platforms, ive districts invest less in students and teachers.

  • The mean spending is different between districts, with districts that are more left leaning, on average, invest more in students and teachers.

  • In addition, the distribution of spending is less ‘bunched’ in liberal Districts a right skew (meaning there are more districts that spend above the national average in students and teachers).

Now, let’s merge Congressional district boundaries and education spending together in a single visualization.

Final Analysis: Mapping Average District Spending by Congressional District

# Subsetting for map layers 
right <- map %>%
  subset(mrp_mean >0)
left <- map %>%
  subset(mrp_mean <0)

# Mapping per pupil Congressional District expenditure
palette <- colorBin("Greens", domain = c(-1.5,7), bins = 9)

leaflet(map) %>%
  addProviderTiles(providers$CartoDB.Positron, group='Political') %>%
  addTiles(group='Physical') %>%
  addPolygons(data=right,
               opacity=0.5,
               weight=2,
               color='salmon',
               fillColor = ~palette(z),
              fillOpacity = 1,
               highlight=highlightOptions(weight=3,
                                        color='aliceblue',
                                        bringToFront=T),
              group="Politically Right") %>%
  addPolygons(data=left,
              opacity=0.5,
              weight=2,
              color='steelblue',
              fillColor = ~palette(z),
              fillOpacity = 1,
              highlight=highlightOptions(weight=3,
                                      color='aliceblue',
                                      bringToFront=T),
               group='Politically Left') %>%
  addLayersControl(baseGroups=c('Political', 'Physical'),
    overlayGroups=c('Politically Right', 'Politically Left'),
    options=layersControlOptions(collapsed = F))%>%
  addLegend('bottomright',
            pal=palette,
            values=~z,
            opacity=0.5,
            title='Per Pupil Expenditure, scaled') 
# Mapping per instructor Congressional District Expenditure
palette_i <- colorBin("Greens", domain = c(-1.3,6), bins = 9)

leaflet(map) %>%
  addProviderTiles(providers$CartoDB.Positron, group='Political') %>%
  addTiles(group='Physical') %>%
  addPolygons(data=right,
              opacity=0.5,
              weight=2,
              color='salmon',
              fillColor = ~palette_i(z_i),
              fillOpacity = 1,
              highlight=highlightOptions(weight=3,
                                        color='aliceblue',
                                        bringToFront=T),
              group="Politically Right") %>%
  addPolygons(data=left,
              opacity=0.5,
              weight=2,
              color='steelblue',
              fillColor = ~palette_i(z_i),
              fillOpacity = 1,
              highlight=highlightOptions(weight=3,
                                      color='aliceblue',
                                      bringToFront=T),
               group='Politically Left') %>%
  addLayersControl(baseGroups=c('Political', 'Physical'),
    overlayGroups=c('Politically Right', 'Politically Left'),
    options=layersControlOptions(collapsed = F))%>%
  addLegend('bottomright',
            pal=palette_i,
            values=~z_i,
            opacity=0.5,
            title='Per Instructor Expenditure, scaled')

Toggling between the map layers provides an interactive visual of the geographic distribution of Congressional districts lines and spending on education by political leaning.

We do indeed find spatial correlations between education spending that aligns with Congressional districts.

In summary,

It is clear, that investing in students and teachers to ensure that all can access the support needed is critical for equity in education. We find that despite that the for right-leaning Districts, the magnitude in which constituent preferences are predominantly ive is much greater than for left-leaning Districts. In other words, Districts that are more left-leaning, are less radically liberal (and more neutral) than right-leaning Districts are radically ive.

Yet, left leaning districts invest more on average in students and teachers, with the leading Districts left-leaning. Further, follow up analyses may want to look at other outcomes, such as student teacher ratio and proportions of special education students, english language learners, special education teachers, and guidance counselors in school districts between Congressional Districts.